peer effects and externalities in technology adoption: evidence from community reporting in uganda
Post on 21-Jan-2018
229 Views
Preview:
TRANSCRIPT
Peer effects and externalities in technology adoption:Evidence from community reporting in Uganda
Romain Ferrali1 Guy Grossman2 Melina Platas Izama3 Jonathan Rodden4
1Princeton 2UPenn 3NYU Abu-Dhabi 4Stanford
December 15, 2017
SITE – Stockholm School of Economics
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 1 / 29
Overview
Motivation
Developing countries: persistent poor public service delivery
← frontlineproviders not sufficiently monitored / supervised by public officials
Possible solution: community monitoring – monitoring costs can be high(time, possible retaliation) while benefits are diffused
Modified solution: ICTs platforms that support community reporting(immediate, inexpensive, anonymous, “comparative advantage”)
Take-up: matters for both efficiency and equity reasons
Variable adoption rates → why do community reporting ICT platforms (andPCTs more generally) get adopted in some places and not in other?
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 2 / 29
Overview
Motivation
Developing countries: persistent poor public service delivery ← frontlineproviders not sufficiently monitored / supervised by public officials
Possible solution: community monitoring – monitoring costs can be high(time, possible retaliation) while benefits are diffused
Modified solution: ICTs platforms that support community reporting(immediate, inexpensive, anonymous, “comparative advantage”)
Take-up: matters for both efficiency and equity reasons
Variable adoption rates → why do community reporting ICT platforms (andPCTs more generally) get adopted in some places and not in other?
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 2 / 29
Overview
Motivation
Developing countries: persistent poor public service delivery ← frontlineproviders not sufficiently monitored / supervised by public officials
Possible solution: community monitoring
– monitoring costs can be high(time, possible retaliation) while benefits are diffused
Modified solution: ICTs platforms that support community reporting(immediate, inexpensive, anonymous, “comparative advantage”)
Take-up: matters for both efficiency and equity reasons
Variable adoption rates → why do community reporting ICT platforms (andPCTs more generally) get adopted in some places and not in other?
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 2 / 29
Overview
Motivation
Developing countries: persistent poor public service delivery ← frontlineproviders not sufficiently monitored / supervised by public officials
Possible solution: community monitoring – monitoring costs can be high(time, possible retaliation) while benefits are diffused
Modified solution: ICTs platforms that support community reporting(immediate, inexpensive, anonymous, “comparative advantage”)
Take-up: matters for both efficiency and equity reasons
Variable adoption rates → why do community reporting ICT platforms (andPCTs more generally) get adopted in some places and not in other?
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 2 / 29
Overview
Motivation
Developing countries: persistent poor public service delivery ← frontlineproviders not sufficiently monitored / supervised by public officials
Possible solution: community monitoring – monitoring costs can be high(time, possible retaliation) while benefits are diffused
Modified solution: ICTs platforms that support community reporting(immediate, inexpensive, anonymous, “comparative advantage”)
Take-up: matters for both efficiency and equity reasons
Variable adoption rates → why do community reporting ICT platforms (andPCTs more generally) get adopted in some places and not in other?
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 2 / 29
Overview
Motivation
Developing countries: persistent poor public service delivery ← frontlineproviders not sufficiently monitored / supervised by public officials
Possible solution: community monitoring – monitoring costs can be high(time, possible retaliation) while benefits are diffused
Modified solution: ICTs platforms that support community reporting(immediate, inexpensive, anonymous, “comparative advantage”)
Take-up: matters for both efficiency and equity reasons
Variable adoption rates → why do community reporting ICT platforms (andPCTs more generally) get adopted in some places and not in other?
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 2 / 29
Overview
Motivation
Developing countries: persistent poor public service delivery ← frontlineproviders not sufficiently monitored / supervised by public officials
Possible solution: community monitoring – monitoring costs can be high(time, possible retaliation) while benefits are diffused
Modified solution: ICTs platforms that support community reporting(immediate, inexpensive, anonymous, “comparative advantage”)
Take-up: matters for both efficiency and equity reasons
Variable adoption rates
→ why do community reporting ICT platforms (andPCTs more generally) get adopted in some places and not in other?
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 2 / 29
Overview
Motivation
Developing countries: persistent poor public service delivery ← frontlineproviders not sufficiently monitored / supervised by public officials
Possible solution: community monitoring – monitoring costs can be high(time, possible retaliation) while benefits are diffused
Modified solution: ICTs platforms that support community reporting(immediate, inexpensive, anonymous, “comparative advantage”)
Take-up: matters for both efficiency and equity reasons
Variable adoption rates → why do community reporting ICT platforms (andPCTs more generally) get adopted in some places and not in other?
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 2 / 29
Overview
Technology adoption literature: networks matter!
Social learning from peers (friends, family, co-workers) facilitates theadoption of new technologies
Early adopting peers (network ties) reduce uncertainty over costs and benefitsof using a new technology
Costs of communication with network ties are lower and their opinion isgenerally more trustworthy
Networks shown to support diffusion of technologies from agriculture bestpractices (Conley & Udry 2010) to medical innovations (Coleman 1966), etc.
Open question: do networks support the adoption of politicalcommunication technologies, such as community reporting platforms?
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 3 / 29
Overview
Technology adoption literature: networks matter!
Social learning from peers (friends, family, co-workers) facilitates theadoption of new technologies
Early adopting peers (network ties) reduce uncertainty over costs and benefitsof using a new technology
Costs of communication with network ties are lower and their opinion isgenerally more trustworthy
Networks shown to support diffusion of technologies from agriculture bestpractices (Conley & Udry 2010) to medical innovations (Coleman 1966), etc.
Open question: do networks support the adoption of politicalcommunication technologies, such as community reporting platforms?
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 3 / 29
Overview
Technology adoption literature: networks matter!
Social learning from peers (friends, family, co-workers) facilitates theadoption of new technologies
Early adopting peers (network ties) reduce uncertainty over costs and benefitsof using a new technology
Costs of communication with network ties are lower and their opinion isgenerally more trustworthy
Networks shown to support diffusion of technologies from agriculture bestpractices (Conley & Udry 2010) to medical innovations (Coleman 1966), etc.
Open question: do networks support the adoption of politicalcommunication technologies, such as community reporting platforms?
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 3 / 29
Overview
Technology adoption literature: networks matter!
Social learning from peers (friends, family, co-workers) facilitates theadoption of new technologies
Early adopting peers (network ties) reduce uncertainty over costs and benefitsof using a new technology
Costs of communication with network ties are lower and their opinion isgenerally more trustworthy
Networks shown to support diffusion of technologies from agriculture bestpractices (Conley & Udry 2010) to medical innovations (Coleman 1966), etc.
Open question: do networks support the adoption of politicalcommunication technologies, such as community reporting platforms?
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 3 / 29
Overview
Technology adoption literature: networks matter!
Social learning from peers (friends, family, co-workers) facilitates theadoption of new technologies
Early adopting peers (network ties) reduce uncertainty over costs and benefitsof using a new technology
Costs of communication with network ties are lower and their opinion isgenerally more trustworthy
Networks shown to support diffusion of technologies from agriculture bestpractices (Conley & Udry 2010) to medical innovations (Coleman 1966), etc.
Open question: do networks support the adoption of politicalcommunication technologies, such as community reporting platforms?
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 3 / 29
Overview
Research design in nutshell
We study how social ties contribute to the adoption of a new mobile-basedpolitical communication platform in a low-income country setting
The communication platform enables citizens to report service deliveryproblems via text-messages to their (Ugandan) local government
The political communication platform was introduced to 130 Ugandanvillages using a field experimental design
Consistent with past research on “ICT for better governance,” adoption ratesamong the treatment villages were highly uneven
We explore the role of social diffusion by collecting ‘whole’ network data from16 treatment villages as well as data on knowledge and usage of the platform
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 4 / 29
Overview
Research design in nutshell
We study how social ties contribute to the adoption of a new mobile-basedpolitical communication platform in a low-income country setting
The communication platform enables citizens to report service deliveryproblems via text-messages to their (Ugandan) local government
The political communication platform was introduced to 130 Ugandanvillages using a field experimental design
Consistent with past research on “ICT for better governance,” adoption ratesamong the treatment villages were highly uneven
We explore the role of social diffusion by collecting ‘whole’ network data from16 treatment villages as well as data on knowledge and usage of the platform
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 4 / 29
Overview
Research design in nutshell
We study how social ties contribute to the adoption of a new mobile-basedpolitical communication platform in a low-income country setting
The communication platform enables citizens to report service deliveryproblems via text-messages to their (Ugandan) local government
The political communication platform was introduced to 130 Ugandanvillages using a field experimental design
Consistent with past research on “ICT for better governance,” adoption ratesamong the treatment villages were highly uneven
We explore the role of social diffusion by collecting ‘whole’ network data from16 treatment villages as well as data on knowledge and usage of the platform
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 4 / 29
Overview
Research design in nutshell
We study how social ties contribute to the adoption of a new mobile-basedpolitical communication platform in a low-income country setting
The communication platform enables citizens to report service deliveryproblems via text-messages to their (Ugandan) local government
The political communication platform was introduced to 130 Ugandanvillages using a field experimental design
Consistent with past research on “ICT for better governance,” adoption ratesamong the treatment villages were highly uneven
We explore the role of social diffusion by collecting ‘whole’ network data from16 treatment villages as well as data on knowledge and usage of the platform
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 4 / 29
Overview
Research design in nutshell
We study how social ties contribute to the adoption of a new mobile-basedpolitical communication platform in a low-income country setting
The communication platform enables citizens to report service deliveryproblems via text-messages to their (Ugandan) local government
The political communication platform was introduced to 130 Ugandanvillages using a field experimental design
Consistent with past research on “ICT for better governance,” adoption ratesamong the treatment villages were highly uneven
We explore the role of social diffusion by collecting ‘whole’ network data from16 treatment villages as well as data on knowledge and usage of the platform
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 4 / 29
Overview
Findings and Theory in Nutshell
1 Empirics I
on average, networks facilitated adoption...
but had no effect in half the villages.
2 Theory: networks effects depend on the goods’ externalities
past work demonstrated peer effects on the adoption of goods with minimalexternalities...
networks have no effect on the adoption of goods characterized by significantexternalities (e.g. PCTs) ...
unless the community enforces truthful communication...
which crucially depends on informal institutions and leadership structure
3 Empirics II
find support for the model’s testable implications
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 5 / 29
Overview
Findings and Theory in Nutshell
1 Empirics I
on average, networks facilitated adoption...
but had no effect in half the villages.
2 Theory: networks effects depend on the goods’ externalities
past work demonstrated peer effects on the adoption of goods with minimalexternalities...
networks have no effect on the adoption of goods characterized by significantexternalities (e.g. PCTs) ...
unless the community enforces truthful communication...
which crucially depends on informal institutions and leadership structure
3 Empirics II
find support for the model’s testable implications
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 5 / 29
Overview
Findings and Theory in Nutshell
1 Empirics I
on average, networks facilitated adoption...
but had no effect in half the villages.
2 Theory: networks effects depend on the goods’ externalities
past work demonstrated peer effects on the adoption of goods with minimalexternalities...
networks have no effect on the adoption of goods characterized by significantexternalities (e.g. PCTs) ...
unless the community enforces truthful communication...
which crucially depends on informal institutions and leadership structure
3 Empirics II
find support for the model’s testable implications
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 5 / 29
Overview
Findings and Theory in Nutshell
1 Empirics I
on average, networks facilitated adoption...
but had no effect in half the villages.
2 Theory: networks effects depend on the goods’ externalities
past work demonstrated peer effects on the adoption of goods with minimalexternalities...
networks have no effect on the adoption of goods characterized by significantexternalities (e.g. PCTs) ...
unless the community enforces truthful communication...
which crucially depends on informal institutions and leadership structure
3 Empirics II
find support for the model’s testable implications
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 5 / 29
Overview
Findings and Theory in Nutshell
1 Empirics I
on average, networks facilitated adoption...
but had no effect in half the villages.
2 Theory: networks effects depend on the goods’ externalities
past work demonstrated peer effects on the adoption of goods with minimalexternalities...
networks have no effect on the adoption of goods characterized by significantexternalities (e.g. PCTs) ...
unless the community enforces truthful communication...
which crucially depends on informal institutions and leadership structure
3 Empirics II
find support for the model’s testable implications
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 5 / 29
Overview
Findings and Theory in Nutshell
1 Empirics I
on average, networks facilitated adoption...
but had no effect in half the villages.
2 Theory: networks effects depend on the goods’ externalities
past work demonstrated peer effects on the adoption of goods with minimalexternalities...
networks have no effect on the adoption of goods characterized by significantexternalities (e.g. PCTs) ...
unless the community enforces truthful communication...
which crucially depends on informal institutions and leadership structure
3 Empirics II
find support for the model’s testable implications
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 5 / 29
Overview
Findings and Theory in Nutshell
1 Empirics I
on average, networks facilitated adoption...
but had no effect in half the villages.
2 Theory: networks effects depend on the goods’ externalities
past work demonstrated peer effects on the adoption of goods with minimalexternalities...
networks have no effect on the adoption of goods characterized by significantexternalities (e.g. PCTs) ...
unless the community enforces truthful communication...
which crucially depends on informal institutions and leadership structure
3 Empirics II
find support for the model’s testable implications
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 5 / 29
Overview
Findings and Theory in Nutshell
1 Empirics I
on average, networks facilitated adoption...
but had no effect in half the villages.
2 Theory: networks effects depend on the goods’ externalities
past work demonstrated peer effects on the adoption of goods with minimalexternalities...
networks have no effect on the adoption of goods characterized by significantexternalities (e.g. PCTs) ...
unless the community enforces truthful communication...
which crucially depends on informal institutions and leadership structure
3 Empirics II
find support for the model’s testable implications
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 5 / 29
Research Design
Research Design
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 6 / 29
Research Design
District local governments in Uganda
Districts: highest tier of subnational government, responsible foradministering local public services (e.g. health, education, water)
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 7 / 29
Research Design
District local governments in Uganda
Districts: highest tier of subnational government, responsible foradministering local public services (e.g. health, education, water)
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 7 / 29
Research Design
Setting
130 randomly selected villages in Arua are encouraged to use platform
U-Bridge effect on service delivery is discussed in a companion paper
Inception community meetings in 24 treatment clusters (October 2014)
Relatively high, but variable technology take-up across villages
Conducted a full census in 16 villages (summer 2016):
8 high performing and 8 low performing wrt uptake (residuals)
Total of 3, 182 villagers
0
2500
5000
7500
10000
12500
2014−07 2015−01 2015−07
Date
Cum
ulat
ive
num
ber
of m
essa
ges
rece
ived
Type
all
relevant
actionable
0
.02
.04
.06
Den
sity
0 20 40 60 80 100relevant messages per 100 villagers
Variability in message intensity
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 8 / 29
Research Design
Setting
130 randomly selected villages in Arua are encouraged to use platform
U-Bridge effect on service delivery is discussed in a companion paper
Inception community meetings in 24 treatment clusters (October 2014)
Relatively high, but variable technology take-up across villages
Conducted a full census in 16 villages (summer 2016):
8 high performing and 8 low performing wrt uptake (residuals)
Total of 3, 182 villagers
0
2500
5000
7500
10000
12500
2014−07 2015−01 2015−07
Date
Cum
ulat
ive
num
ber
of m
essa
ges
rece
ived
Type
all
relevant
actionable
0
.02
.04
.06
Den
sity
0 20 40 60 80 100relevant messages per 100 villagers
Variability in message intensity
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 8 / 29
Research Design
Setting
130 randomly selected villages in Arua are encouraged to use platform
U-Bridge effect on service delivery is discussed in a companion paper
Inception community meetings in 24 treatment clusters (October 2014)
Relatively high, but variable technology take-up across villages
Conducted a full census in 16 villages (summer 2016):
8 high performing and 8 low performing wrt uptake (residuals)
Total of 3, 182 villagers
0
2500
5000
7500
10000
12500
2014−07 2015−01 2015−07
Date
Cum
ulat
ive
num
ber
of m
essa
ges
rece
ived
Type
all
relevant
actionable
0
.02
.04
.06
Den
sity
0 20 40 60 80 100relevant messages per 100 villagers
Variability in message intensity
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 8 / 29
Research Design
Setting
130 randomly selected villages in Arua are encouraged to use platform
U-Bridge effect on service delivery is discussed in a companion paper
Inception community meetings in 24 treatment clusters (October 2014)
Relatively high, but variable technology take-up across villages
Conducted a full census in 16 villages (summer 2016):
8 high performing and 8 low performing wrt uptake (residuals)
Total of 3, 182 villagers
0
2500
5000
7500
10000
12500
2014−07 2015−01 2015−07
Date
Cum
ulat
ive
num
ber
of m
essa
ges
rece
ived
Type
all
relevant
actionable
0
.02
.04
.06
Den
sity
0 20 40 60 80 100relevant messages per 100 villagers
Variability in message intensity
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 8 / 29
Research Design
Setting
130 randomly selected villages in Arua are encouraged to use platform
U-Bridge effect on service delivery is discussed in a companion paper
Inception community meetings in 24 treatment clusters (October 2014)
Relatively high, but variable technology take-up across villages
Conducted a full census in 16 villages (summer 2016):
8 high performing and 8 low performing wrt uptake (residuals)
Total of 3, 182 villagers
0
2500
5000
7500
10000
12500
2014−07 2015−01 2015−07
Date
Cum
ulat
ive
num
ber
of m
essa
ges
rece
ived
Type
all
relevant
actionable
0
.02
.04
.06
Den
sity
0 20 40 60 80 100relevant messages per 100 villagers
Variability in message intensity
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 8 / 29
Research Design
Setting
130 randomly selected villages in Arua are encouraged to use platform
U-Bridge effect on service delivery is discussed in a companion paper
Inception community meetings in 24 treatment clusters (October 2014)
Relatively high, but variable technology take-up across villages
Conducted a full census in 16 villages (summer 2016):
8 high performing and 8 low performing wrt uptake (residuals)
Total of 3, 182 villagers
0
2500
5000
7500
10000
12500
2014−07 2015−01 2015−07
Date
Cum
ulat
ive
num
ber
of m
essa
ges
rece
ived
Type
all
relevant
actionable
0
.02
.04
.06
Den
sity
0 20 40 60 80 100relevant messages per 100 villagers
Variability in message intensity
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 8 / 29
Research Design
Setting
130 randomly selected villages in Arua are encouraged to use platform
U-Bridge effect on service delivery is discussed in a companion paper
Inception community meetings in 24 treatment clusters (October 2014)
Relatively high, but variable technology take-up across villages
Conducted a full census in 16 villages (summer 2016):
8 high performing and 8 low performing wrt uptake (residuals)
Total of 3, 182 villagers
0
2500
5000
7500
10000
12500
2014−07 2015−01 2015−07
Date
Cum
ulat
ive
num
ber
of m
essa
ges
rece
ived
Type
all
relevant
actionable
0
.02
.04
.06
Den
sity
0 20 40 60 80 100relevant messages per 100 villagers
Variability in message intensity
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 8 / 29
Research Design
Setting
130 randomly selected villages in Arua are encouraged to use platform
U-Bridge effect on service delivery is discussed in a companion paper
Inception community meetings in 24 treatment clusters (October 2014)
Relatively high, but variable technology take-up across villages
Conducted a full census in 16 villages (summer 2016):
8 high performing and 8 low performing wrt uptake (residuals)
Total of 3, 182 villagers0
2500
5000
7500
10000
12500
2014−07 2015−01 2015−07
Date
Cum
ulat
ive
num
ber
of m
essa
ges
rece
ived
Type
all
relevant
actionable
0
.02
.04
.06
Den
sity
0 20 40 60 80 100relevant messages per 100 villagers
Variability in message intensity
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 8 / 29
Research Design
Setting
130 randomly selected villages in Arua are encouraged to use platform
U-Bridge effect on service delivery is discussed in a companion paper
Inception community meetings in 24 treatment clusters (October 2014)
Relatively high, but variable technology take-up across villages
Conducted a full census in 16 villages (summer 2016):
8 high performing and 8 low performing wrt uptake (residuals)
Total of 3, 182 villagers
0
2500
5000
7500
10000
12500
2014−07 2015−01 2015−07
Date
Cum
ulat
ive
num
ber
of m
essa
ges
rece
ived
Type
all
relevant
actionable
0
.02
.04
.06
Den
sity
0 20 40 60 80 100relevant messages per 100 villagers
Variability in message intensity
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 8 / 29
Research Design
Constructing the network: four types of ties
Four undirected networks: tie if i names j and j names i
Family
Friends
Lender
Problem solver
Undirected, weighted union network
tie if tie in any of the four networks
weight is number of ties in the four networks
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 9 / 29
Research Design
Constructing the network: four types of ties
Four undirected networks: tie if i names j and j names i
Family
Friends
Lender
Problem solver
Undirected, weighted union network
tie if tie in any of the four networks
weight is number of ties in the four networks
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 9 / 29
Research Design
Figure: Graphical representation of the union network of two villages in the study area.Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 10 / 29
Research Design
Descriptive statistics of sample
Variable Sample Low uptake High uptake ∆ min maxOutcome % adopters 0.04 0.02 0.07 0.05∗∗∗ 0.00 1.00
% heard 0.31 0.23 0.38 0.14∗∗∗ 0.00 1.00% satisfied 0.39 0.22 0.44 0.22∗∗ 0.00 1.00
Individual age 37.39 37.22 37.55 0.33 18 101% females 0.58 0.59 0.56 -0.03∗∗ 0.00 1.00income -0.45 -0.54 -0.36 0.19∗ -2.00 2.00secondary education 0.23 0.18 0.28 0.09∗∗ 0.00 1.00% use phone 0.62 0.58 0.66 0.08∗ 0.00 1.00% immigrants 0.49 0.48 0.50 0.02 0.00 1.00% leaders 0.14 0.12 0.16 0.04∗∗ 0.00 1.00political participation 0.00 -0.00 0.00 0.00 -1.23 1.77% attend meeting 0.08 0.05 0.11 0.06∗∗∗ 0.00 1.00mean pro-sociality 0.28 0.28 0.29 0.01 0.00 1.00
Network degree 8.79 8.36 9.22 0.86 0.00 217.00betweenness 140.60 132.16 149.01 16.85 0.00 23850clustering coefficient 0.38 0.40 0.37 -0.03 0.00 1.00mean size 199.00 198.62 199.38 0.75
Village adult population 269.38 264.25 274.50 10.25 32 429ethnic fractionalization 0.04 0.02 0.07 0.05 0.00 0.41% employed 0.86 0.89 0.84 -0.05 0.68 1.00% non-agriculture 0.22 0.19 0.25 0.06 0.00 0.57poverty score -0.07 -0.09 -0.05 0.03 -0.48 0.47
N 3184 1589 1595 6
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 11 / 29
Research Design
Estimation
Main specification. Spatial Autoregressive Regression (SAR)
y = λMy + Xβ + ε
y , vector of outcomes: adopt ∈ {0, 1}
M, spatial matrix: union network → # adopting neighborsX , matrix of controls
Network: degree centralityDemographics: age, gender, secondary education, immigrantDesign: usage of phone, meeting attendancePolitics: political participation, leadership positionSpatial influence: autoregressive term with M inverse log-distanceWithin-village comparison: models include village-level FE
Robustness checks
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 12 / 29
Research Design
Estimation
Main specification. Spatial Autoregressive Regression (SAR)
y = λMy + Xβ + ε
y , vector of outcomes: adopt ∈ {0, 1}
M, spatial matrix: union network → # adopting neighborsX , matrix of controls
Network: degree centralityDemographics: age, gender, secondary education, immigrantDesign: usage of phone, meeting attendancePolitics: political participation, leadership positionSpatial influence: autoregressive term with M inverse log-distanceWithin-village comparison: models include village-level FE
Robustness checks
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 12 / 29
Research Design
Estimation
Main specification. Spatial Autoregressive Regression (SAR)
y = λMy + Xβ + ε
y , vector of outcomes: adopt ∈ {0, 1}
M, spatial matrix: union network → # adopting neighbors
X , matrix of controlsNetwork: degree centralityDemographics: age, gender, secondary education, immigrantDesign: usage of phone, meeting attendancePolitics: political participation, leadership positionSpatial influence: autoregressive term with M inverse log-distanceWithin-village comparison: models include village-level FE
Robustness checks
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 12 / 29
Research Design
Estimation
Main specification. Spatial Autoregressive Regression (SAR)
y = λMy + Xβ + ε
y , vector of outcomes: adopt ∈ {0, 1}
M, spatial matrix: union network → # adopting neighborsX , matrix of controls
Network: degree centralityDemographics: age, gender, secondary education, immigrantDesign: usage of phone, meeting attendancePolitics: political participation, leadership positionSpatial influence: autoregressive term with M inverse log-distanceWithin-village comparison: models include village-level FE
Robustness checks
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 12 / 29
Research Design
Estimation
Main specification. Spatial Autoregressive Regression (SAR)
y = λMy + Xβ + ε
y , vector of outcomes: adopt ∈ {0, 1}
M, spatial matrix: union network → # adopting neighborsX , matrix of controls
Network: degree centralityDemographics: age, gender, secondary education, immigrantDesign: usage of phone, meeting attendancePolitics: political participation, leadership positionSpatial influence: autoregressive term with M inverse log-distanceWithin-village comparison: models include village-level FE
Robustness checks
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 12 / 29
Main results
Main results
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 13 / 29
Main results
There is peer influence
Dependent variable: adoptAbsolute threshold Fractional threshold
Parsimonious Baseline Parsimonious Baseline(1) (2) (3) (4)
# adopting neighbors 0.035∗∗∗ 0.027∗∗∗
(0.005) (0.005)% adopting neighbors 0.325∗∗∗ 0.213∗∗∗
(0.052) (0.048)degree 0.002∗∗∗ 0.001∗ 0.004∗∗∗ 0.003∗∗∗
(0.001) (0.001) (0.001) (0.001)Village FE X X X XControls X XObservations 3,184 3,019 3,184 3,019R2 0.139 0.245 0.116 0.231
Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 14 / 29
Main results
Robustness checks
Threats to identification:1 Networks may matter differently for “hearing” and “adopting”2 Peer effects may be spurious (homophily, shared context, ...)3 Exposure to encouragements is endogenous to network position
Test Issue Result NotesSelection modelPr(adopt) = Pr(hear)×Pr(adopt|hear)(Larson & Lewis 2017)
1 X peers affect both stages of diffu-sion; adoption variability larger
Instrumental variable(An 2016)
2 X zj → yj → yiInstrument: distance from meetinglocation
Non-parametric controls for degree(Aronow & Samii, nd)
3 X degree strata & GAM
Matching(Aral et al 2009)
2, 3 X full matching on network covariatesand most important predictors ofuptake
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 15 / 29
Main results
Robustness checks
Threats to identification:1 Networks may matter differently for “hearing” and “adopting”2 Peer effects may be spurious (homophily, shared context, ...)3 Exposure to encouragements is endogenous to network position
Test Issue Result NotesSelection modelPr(adopt) = Pr(hear)×Pr(adopt|hear)(Larson & Lewis 2017)
1 X peers affect both stages of diffu-sion; adoption variability larger
Instrumental variable(An 2016)
2 X zj → yj → yiInstrument: distance from meetinglocation
Non-parametric controls for degree(Aronow & Samii, nd)
3 X degree strata & GAM
Matching(Aral et al 2009)
2, 3 X full matching on network covariatesand most important predictors ofuptake
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 15 / 29
Main results
Robustness checks
Threats to identification:1 Networks may matter differently for “hearing” and “adopting”2 Peer effects may be spurious (homophily, shared context, ...)3 Exposure to encouragements is endogenous to network position
Test Issue Result NotesSelection modelPr(adopt) = Pr(hear)×Pr(adopt|hear)(Larson & Lewis 2017)
1 X peers affect both stages of diffu-sion; adoption variability larger
Instrumental variable(An 2016)
2 X zj → yj → yiInstrument: distance from meetinglocation
Non-parametric controls for degree(Aronow & Samii, nd)
3 X degree strata & GAM
Matching(Aral et al 2009)
2, 3 X full matching on network covariatesand most important predictors ofuptake
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 15 / 29
Main results
Robustness checks
Threats to identification:1 Networks may matter differently for “hearing” and “adopting”2 Peer effects may be spurious (homophily, shared context, ...)3 Exposure to encouragements is endogenous to network position
Test Issue Result NotesSelection modelPr(adopt) = Pr(hear)×Pr(adopt|hear)(Larson & Lewis 2017)
1 X peers affect both stages of diffu-sion; adoption variability larger
Instrumental variable(An 2016)
2 X zj → yj → yiInstrument: distance from meetinglocation
Non-parametric controls for degree(Aronow & Samii, nd)
3 X degree strata & GAM
Matching(Aral et al 2009)
2, 3 X full matching on network covariatesand most important predictors ofuptake
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 15 / 29
Main results
Robustness checks
Threats to identification:1 Networks may matter differently for “hearing” and “adopting”2 Peer effects may be spurious (homophily, shared context, ...)3 Exposure to encouragements is endogenous to network position
Test Issue Result NotesSelection modelPr(adopt) = Pr(hear)×Pr(adopt|hear)(Larson & Lewis 2017)
1 X peers affect both stages of diffu-sion; adoption variability larger
Instrumental variable(An 2016)
2 X zj → yj → yiInstrument: distance from meetinglocation
Non-parametric controls for degree(Aronow & Samii, nd)
3 X degree strata & GAM
Matching(Aral et al 2009)
2, 3 X full matching on network covariatesand most important predictors ofuptake
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 15 / 29
Main results
Wide variation across villages
●
●●
●
●
●●
−0.05
0.00
0.05
0.10
B(236)
H(102)
E(263)
C(159)
M(195)
G(163)
D(282)
K(204)
L(228)
P(191)
O(187)
I(168)
F(205)
N(223)
J(185)
Village
Ave
rage
Mar
gina
l Effe
ct
Uptake
● High
Low
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 16 / 29
Model
Model
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 17 / 29
Model
Theory
Adopting a new technology is risky
Potential adopters rely on peers to learn about the costs and benefits
Lying about benefits of technology is costly
Technologies vary in whether adoption generates externalities:
Externalities in adoption: whether the adoption decision of one agentaffects the payoff from adoption of another agent
No externalities (private good). truthful communication is an equilibrium ⇒networks foster learning
Positive externalities (political participation). no truthful communication ⇒networks are ineffective... unless the community can enforce truthfulcommunication.
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 18 / 29
Model
Theory
Adopting a new technology is risky
Potential adopters rely on peers to learn about the costs and benefits
Lying about benefits of technology is costly
Technologies vary in whether adoption generates externalities:
Externalities in adoption: whether the adoption decision of one agentaffects the payoff from adoption of another agent
No externalities (private good). truthful communication is an equilibrium ⇒networks foster learning
Positive externalities (political participation). no truthful communication ⇒networks are ineffective... unless the community can enforce truthfulcommunication.
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 18 / 29
Model
Theory
Adopting a new technology is risky
Potential adopters rely on peers to learn about the costs and benefits
Lying about benefits of technology is costly
Technologies vary in whether adoption generates externalities:
Externalities in adoption: whether the adoption decision of one agentaffects the payoff from adoption of another agent
No externalities (private good). truthful communication is an equilibrium ⇒networks foster learning
Positive externalities (political participation). no truthful communication ⇒networks are ineffective... unless the community can enforce truthfulcommunication.
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 18 / 29
Model
Theory
Adopting a new technology is risky
Potential adopters rely on peers to learn about the costs and benefits
Lying about benefits of technology is costly
Technologies vary in whether adoption generates externalities:
Externalities in adoption: whether the adoption decision of one agentaffects the payoff from adoption of another agent
No externalities (private good). truthful communication is an equilibrium ⇒networks foster learning
Positive externalities (political participation). no truthful communication ⇒networks are ineffective... unless the community can enforce truthfulcommunication.
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 18 / 29
Model
Theory
Adopting a new technology is risky
Potential adopters rely on peers to learn about the costs and benefits
Lying about benefits of technology is costly
Technologies vary in whether adoption generates externalities:
Externalities in adoption: whether the adoption decision of one agentaffects the payoff from adoption of another agent
No externalities (private good). truthful communication is an equilibrium ⇒networks foster learning
Positive externalities (political participation). no truthful communication ⇒networks are ineffective... unless the community can enforce truthfulcommunication.
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 18 / 29
Model
Theory
Adopting a new technology is risky
Potential adopters rely on peers to learn about the costs and benefits
Lying about benefits of technology is costly
Technologies vary in whether adoption generates externalities:
Externalities in adoption: whether the adoption decision of one agentaffects the payoff from adoption of another agent
No externalities (private good). truthful communication is an equilibrium ⇒networks foster learning
Positive externalities (political participation). no truthful communication ⇒networks are ineffective... unless the community can enforce truthfulcommunication.
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 18 / 29
Model
Theory
Adopting a new technology is risky
Potential adopters rely on peers to learn about the costs and benefits
Lying about benefits of technology is costly
Technologies vary in whether adoption generates externalities:
Externalities in adoption: whether the adoption decision of one agentaffects the payoff from adoption of another agent
No externalities (private good). truthful communication is an equilibrium ⇒networks foster learning
Positive externalities (political participation). no truthful communication ⇒networks are ineffective...
unless the community can enforce truthfulcommunication.
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 18 / 29
Model
Theory
Adopting a new technology is risky
Potential adopters rely on peers to learn about the costs and benefits
Lying about benefits of technology is costly
Technologies vary in whether adoption generates externalities:
Externalities in adoption: whether the adoption decision of one agentaffects the payoff from adoption of another agent
No externalities (private good). truthful communication is an equilibrium ⇒networks foster learning
Positive externalities (political participation). no truthful communication ⇒networks are ineffective... unless the community can enforce truthfulcommunication.
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 18 / 29
Model
A model: adoption without externalities
N agents are the nodes of a social network gEach agent i decides whether to adopt a new technology, yi ∈ {0, 1}.
ui (yi , θ) = qθ(yi )− yici
Not adopting gives a payoff of zero: qθ(0) = 0Adoption is costly: ci ∈ (0, 1)Adoption is risky:
at t = 0, nature draws state of the world θ ∈ {H, L}.i is more likely gets benefit B = 1 in the high state: qH(1) > qL(1) = 0
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 19 / 29
Model
Learning and communication
t = 0: nature draws the state θ ∈ {H, L}
t = 1: each agent gets a signal si ∈ {H, L} that matches the state withprobability pi .
pi ≡ expertiset = 2: communication. Each agent i sends a message mij ∈ {H, L} abouttheir signal to each of their neighbors. j ∈ Ni (g).t = 3: agents update their belief about the state, and decide whether toadopt if sufficiently confident they are in high state
yi = 1 ⇐⇒ Pr(θ = H|si , {mji})Pr(θ = L|si , {mji})︸ ︷︷ ︸
likelihood ratio
≥ ai︸︷︷︸ = f (ci , πi )
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 20 / 29
Model
Learning and communication
t = 0: nature draws the state θ ∈ {H, L}t = 1: each agent gets a signal si ∈ {H, L} that matches the state withprobability pi .
pi ≡ expertiset = 2: communication. Each agent i sends a message mij ∈ {H, L} abouttheir signal to each of their neighbors. j ∈ Ni (g).t = 3: agents update their belief about the state, and decide whether toadopt if sufficiently confident they are in high state
yi = 1 ⇐⇒ Pr(θ = H|si , {mji})Pr(θ = L|si , {mji})︸ ︷︷ ︸
likelihood ratio
≥ ai︸︷︷︸ = f (ci , πi )
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 20 / 29
Model
Learning and communication
t = 0: nature draws the state θ ∈ {H, L}t = 1: each agent gets a signal si ∈ {H, L} that matches the state withprobability pi .
pi ≡ expertise
t = 2: communication. Each agent i sends a message mij ∈ {H, L} abouttheir signal to each of their neighbors. j ∈ Ni (g).t = 3: agents update their belief about the state, and decide whether toadopt if sufficiently confident they are in high state
yi = 1 ⇐⇒ Pr(θ = H|si , {mji})Pr(θ = L|si , {mji})︸ ︷︷ ︸
likelihood ratio
≥ ai︸︷︷︸ = f (ci , πi )
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 20 / 29
Model
Learning and communication
t = 0: nature draws the state θ ∈ {H, L}t = 1: each agent gets a signal si ∈ {H, L} that matches the state withprobability pi .
pi ≡ expertiset = 2: communication. Each agent i sends a message mij ∈ {H, L} abouttheir signal to each of their neighbors. j ∈ Ni (g).
t = 3: agents update their belief about the state, and decide whether toadopt if sufficiently confident they are in high state
yi = 1 ⇐⇒ Pr(θ = H|si , {mji})Pr(θ = L|si , {mji})︸ ︷︷ ︸
likelihood ratio
≥ ai︸︷︷︸ = f (ci , πi )
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 20 / 29
Model
Learning and communication
t = 0: nature draws the state θ ∈ {H, L}t = 1: each agent gets a signal si ∈ {H, L} that matches the state withprobability pi .
pi ≡ expertiset = 2: communication. Each agent i sends a message mij ∈ {H, L} abouttheir signal to each of their neighbors. j ∈ Ni (g).t = 3: agents update their belief about the state, and decide whether toadopt if sufficiently confident they are in high state
yi = 1 ⇐⇒ Pr(θ = H|si , {mji})Pr(θ = L|si , {mji})︸ ︷︷ ︸
likelihood ratio
≥ ai︸︷︷︸ = f (ci , πi )
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 20 / 29
Model
Learning and communication
t = 0: nature draws the state θ ∈ {H, L}t = 1: each agent gets a signal si ∈ {H, L} that matches the state withprobability pi .
pi ≡ expertiset = 2: communication. Each agent i sends a message mij ∈ {H, L} abouttheir signal to each of their neighbors. j ∈ Ni (g).t = 3: agents update their belief about the state, and decide whether toadopt if sufficiently confident they are in high state
yi = 1 ⇐⇒ Pr(θ = H|si , {mji})Pr(θ = L|si , {mji})︸ ︷︷ ︸
likelihood ratio
≥ ai︸︷︷︸
= f (ci , πi )
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 20 / 29
Model
Learning and communication
t = 0: nature draws the state θ ∈ {H, L}t = 1: each agent gets a signal si ∈ {H, L} that matches the state withprobability pi .
pi ≡ expertiset = 2: communication. Each agent i sends a message mij ∈ {H, L} abouttheir signal to each of their neighbors. j ∈ Ni (g).t = 3: agents update their belief about the state, and decide whether toadopt if sufficiently confident they are in high state
yi = 1 ⇐⇒ Pr(θ = H|si , {mji})Pr(θ = L|si , {mji})︸ ︷︷ ︸
likelihood ratio
≥ ai︸︷︷︸ = f (ci , πi )
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 20 / 29
Model
The benefits of truthful communication
Truthful communication fosters learning:
1 More peers ⇒ better learning
2 Outcomes of peers are correlated
3 Agents put more weight on experts
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 21 / 29
Model
When do you get truthful communication? (Setup)
The case without externalities
ui ={
ui (yi , θ) = qθ(yi )− yici , without externalitiesui (yi , y−i , θ) = qθ
(yi +
∑j 6=i yj
)− yici , with positive externalities
Additional assumption: qL(y) = 0 ≤ qH(y) ≤ qH(y + 1).Introducing a cost of lying κ ≥ 0:
ui = ui (yi , ., θ,mi ) = ui (yi , ., θ)− κ∑j 6=i
1{mij 6= si}
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 22 / 29
Model
When do you get truthful communication? (Setup)
The case without externalities
ui ={
ui (yi , θ) = qθ(yi )− yici , without externalitiesui (yi , y−i , θ) = qθ
(yi +
∑j 6=i yj
)− yici , with positive externalities
Additional assumption:
qL(y) = 0 ≤ qH(y) ≤ qH(y + 1).Introducing a cost of lying κ ≥ 0:
ui = ui (yi , ., θ,mi ) = ui (yi , ., θ)− κ∑j 6=i
1{mij 6= si}
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 22 / 29
Model
When do you get truthful communication? (Setup)
The case without externalities
ui ={
ui (yi , θ) = qθ(yi )− yici , without externalitiesui (yi , y−i , θ) = qθ
(yi +
∑j 6=i yj
)− yici , with positive externalities
Additional assumption: qL(y) = 0 ≤ qH(y) ≤ qH(y + 1).Introducing a cost of lying κ ≥ 0:
ui = ui (yi , ., θ,mi ) = ui (yi , ., θ)− κ∑j 6=i
1{mij 6= si}
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 22 / 29
Model
When do you get truthful communication? (Setup)
The case without externalities
ui ={
ui (yi , θ) = qθ(yi )− yici , without externalitiesui (yi , y−i , θ) = qθ
(yi +
∑j 6=i yj
)− yici , with positive externalities
Additional assumption: qL(y) = 0 ≤ qH(y) ≤ qH(y + 1).Introducing a cost of lying κ ≥ 0:
ui = ui (yi , ., θ,mi ) = ui (yi , ., θ)− κ∑j 6=i
1{mij 6= si}
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 22 / 29
Model
When do you get truthful communication?
No externalities:Lying brings no benefits and generates costsTC is an equilibrium for any κ ≥ 0TC is the unique equilibrium for any κ > 0
Positive externalities:lying brings benefit and TC is not equilibriumpeer effects depend on making cost of lying high enoughTC is an equilibrium iff κ ≥ κ̄1 ← informal institutions!TC is the unique equilibrium iff κ ≥ κ̄20 ≤ κ̄1 ≤ κ̄2 ≤ 1
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 23 / 29
Model
When do you get truthful communication?
No externalities:Lying brings no benefits and generates costsTC is an equilibrium for any κ ≥ 0TC is the unique equilibrium for any κ > 0
Positive externalities:lying brings benefit and TC is not equilibriumpeer effects depend on making cost of lying high enoughTC is an equilibrium iff κ ≥ κ̄1 ← informal institutions!TC is the unique equilibrium iff κ ≥ κ̄20 ≤ κ̄1 ≤ κ̄2 ≤ 1
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 23 / 29
Empirical implications
Empirical implications
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 24 / 29
Empirical implications
Empirical implications
1 Variation across networks in the support of diffusion of goods withexternalities, above what can be explained by variation in hearing rates
2 Discounting of positive signals (peers’ recommendations) when truthfulcommunication is not enforced
3 Experts will have a stronger peer effect than novices iff a network supportsdiffusion, as their signal carries greater weight
4 Strong ties will be more effective than weak ties in supporting truthfulcommunication, and therefore, in supporting diffusion
5 Informal institutions should support adoption in high-uptake villages
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 25 / 29
Empirical implications
Empirical implications
1 Variation across networks in the support of diffusion of goods withexternalities, above what can be explained by variation in hearing rates
2 Discounting of positive signals (peers’ recommendations) when truthfulcommunication is not enforced
3 Experts will have a stronger peer effect than novices iff a network supportsdiffusion, as their signal carries greater weight
4 Strong ties will be more effective than weak ties in supporting truthfulcommunication, and therefore, in supporting diffusion
5 Informal institutions should support adoption in high-uptake villages
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 25 / 29
Empirical implications
Empirical implications
1 Variation across networks in the support of diffusion of goods withexternalities, above what can be explained by variation in hearing rates
2 Discounting of positive signals (peers’ recommendations) when truthfulcommunication is not enforced
3 Experts will have a stronger peer effect than novices iff a network supportsdiffusion, as their signal carries greater weight
4 Strong ties will be more effective than weak ties in supporting truthfulcommunication, and therefore, in supporting diffusion
5 Informal institutions should support adoption in high-uptake villages
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 25 / 29
Empirical implications
Empirical implications
1 Variation across networks in the support of diffusion of goods withexternalities, above what can be explained by variation in hearing rates
2 Discounting of positive signals (peers’ recommendations) when truthfulcommunication is not enforced
3 Experts will have a stronger peer effect than novices iff a network supportsdiffusion, as their signal carries greater weight
4 Strong ties will be more effective than weak ties in supporting truthfulcommunication, and therefore, in supporting diffusion
5 Informal institutions should support adoption in high-uptake villages
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 25 / 29
Empirical implications
Empirical implications
1 Variation across networks in the support of diffusion of goods withexternalities, above what can be explained by variation in hearing rates
2 Discounting of positive signals (peers’ recommendations) when truthfulcommunication is not enforced
3 Experts will have a stronger peer effect than novices iff a network supportsdiffusion, as their signal carries greater weight
4 Strong ties will be more effective than weak ties in supporting truthfulcommunication, and therefore, in supporting diffusion
5 Informal institutions should support adoption in high-uptake villages
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 25 / 29
Empirical implications
Findings
1 Variation across networks in the support of diffusion of goods withexternalities
There are peer effects, but not in all villages.
2 Discounting of positive signals (peers’ recommendations) when truthfulcommunication is not enforced
i is more likely to adopt if j is satisfied in high uptake villages
3 Experts will have a stronger peer effect than novices
Leaders exert more influence than citizens in high-uptake villages
4 Informal institutions should support adoption in high-uptake villages
Public goods games and concentrated leadership
●
●●
●
●
●●
−0.05
0.00
0.05
0.10
B(236)
H(102)
E(263)
C(159)
M(195)
G(163)
D(282)
K(204)
L(228)
P(191)
O(187)
I(168)
F(205)
N(223)
J(185)
Village
Ave
rage
Mar
gina
l Effe
ct
Uptake
● High
Low
High uptakevillage
Low uptakevillage
0.00 0.03 0.06 0.09
Average marginal effect of one adopting neighbor on adoption
Effect
satisfaction
communication
contagion
●
●
●
●
Low uptake
High uptake
−0.01 0.00 0.01 0.02 0.03 0.04 0.05
Leader
Peer
Leader
Peer
Average Marginal Effect
sour
ce
●
●
●
●
●
●
●
●
●
●
Ethnic concentration (12)
Pct. strong ties (15)
Pct. leaders (15)
Degree (15)
Diffusion potential (15)
Population (15)
Religious concentration (15)
Pro−sociality − dictator (15)
Pro−sociality − public good (15)
Leadership concentration (14)
−5.0 −2.5 0.0 2.5
Standardized effect size
Var
iabl
e
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 26 / 29
Empirical implications
Findings
1 Variation across networks in the support of diffusion of goods withexternalities
There are peer effects, but not in all villages.
2 Discounting of positive signals (peers’ recommendations) when truthfulcommunication is not enforced
i is more likely to adopt if j is satisfied in high uptake villages
3 Experts will have a stronger peer effect than novices
Leaders exert more influence than citizens in high-uptake villages
4 Informal institutions should support adoption in high-uptake villages
Public goods games and concentrated leadership
●
●●
●
●
●●
−0.05
0.00
0.05
0.10
B(236)
H(102)
E(263)
C(159)
M(195)
G(163)
D(282)
K(204)
L(228)
P(191)
O(187)
I(168)
F(205)
N(223)
J(185)
Village
Ave
rage
Mar
gina
l Effe
ct
Uptake
● High
Low
High uptakevillage
Low uptakevillage
0.00 0.03 0.06 0.09
Average marginal effect of one adopting neighbor on adoption
Effect
satisfaction
communication
contagion
●
●
●
●
Low uptake
High uptake
−0.01 0.00 0.01 0.02 0.03 0.04 0.05
Leader
Peer
Leader
Peer
Average Marginal Effect
sour
ce
●
●
●
●
●
●
●
●
●
●
Ethnic concentration (12)
Pct. strong ties (15)
Pct. leaders (15)
Degree (15)
Diffusion potential (15)
Population (15)
Religious concentration (15)
Pro−sociality − dictator (15)
Pro−sociality − public good (15)
Leadership concentration (14)
−5.0 −2.5 0.0 2.5
Standardized effect size
Var
iabl
e
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 26 / 29
Empirical implications
Findings
1 Variation across networks in the support of diffusion of goods withexternalities
There are peer effects, but not in all villages.
2 Discounting of positive signals (peers’ recommendations) when truthfulcommunication is not enforced
i is more likely to adopt if j is satisfied in high uptake villages
3 Experts will have a stronger peer effect than novices
Leaders exert more influence than citizens in high-uptake villages
4 Informal institutions should support adoption in high-uptake villages
Public goods games and concentrated leadership
●
●●
●
●
●●
−0.05
0.00
0.05
0.10
B(236)
H(102)
E(263)
C(159)
M(195)
G(163)
D(282)
K(204)
L(228)
P(191)
O(187)
I(168)
F(205)
N(223)
J(185)
Village
Ave
rage
Mar
gina
l Effe
ct
Uptake
● High
Low
High uptakevillage
Low uptakevillage
0.00 0.03 0.06 0.09
Average marginal effect of one adopting neighbor on adoption
Effect
satisfaction
communication
contagion
●
●
●
●
Low uptake
High uptake
−0.01 0.00 0.01 0.02 0.03 0.04 0.05
Leader
Peer
Leader
Peer
Average Marginal Effect
sour
ce
●
●
●
●
●
●
●
●
●
●
Ethnic concentration (12)
Pct. strong ties (15)
Pct. leaders (15)
Degree (15)
Diffusion potential (15)
Population (15)
Religious concentration (15)
Pro−sociality − dictator (15)
Pro−sociality − public good (15)
Leadership concentration (14)
−5.0 −2.5 0.0 2.5
Standardized effect size
Var
iabl
e
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 26 / 29
Empirical implications
Findings
1 Variation across networks in the support of diffusion of goods withexternalities
There are peer effects, but not in all villages.
2 Discounting of positive signals (peers’ recommendations) when truthfulcommunication is not enforced
i is more likely to adopt if j is satisfied in high uptake villages
3 Experts will have a stronger peer effect than novices
Leaders exert more influence than citizens in high-uptake villages
4 Informal institutions should support adoption in high-uptake villages
Public goods games and concentrated leadership
●
●●
●
●
●●
−0.05
0.00
0.05
0.10
B(236)
H(102)
E(263)
C(159)
M(195)
G(163)
D(282)
K(204)
L(228)
P(191)
O(187)
I(168)
F(205)
N(223)
J(185)
Village
Ave
rage
Mar
gina
l Effe
ct
Uptake
● High
Low
High uptakevillage
Low uptakevillage
0.00 0.03 0.06 0.09
Average marginal effect of one adopting neighbor on adoption
Effect
satisfaction
communication
contagion
●
●
●
●
Low uptake
High uptake
−0.01 0.00 0.01 0.02 0.03 0.04 0.05
Leader
Peer
Leader
Peer
Average Marginal Effect
sour
ce
●
●
●
●
●
●
●
●
●
●
Ethnic concentration (12)
Pct. strong ties (15)
Pct. leaders (15)
Degree (15)
Diffusion potential (15)
Population (15)
Religious concentration (15)
Pro−sociality − dictator (15)
Pro−sociality − public good (15)
Leadership concentration (14)
−5.0 −2.5 0.0 2.5
Standardized effect size
Var
iabl
e
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 26 / 29
Empirical implications
Findings
1 Variation across networks in the support of diffusion of goods withexternalities
There are peer effects, but not in all villages.
2 Discounting of positive signals (peers’ recommendations) when truthfulcommunication is not enforced
i is more likely to adopt if j is satisfied in high uptake villages
3 Experts will have a stronger peer effect than novices
Leaders exert more influence than citizens in high-uptake villages
4 Informal institutions should support adoption in high-uptake villages
Public goods games and concentrated leadership
●
●●
●
●
●●
−0.05
0.00
0.05
0.10
B(236)
H(102)
E(263)
C(159)
M(195)
G(163)
D(282)
K(204)
L(228)
P(191)
O(187)
I(168)
F(205)
N(223)
J(185)
Village
Ave
rage
Mar
gina
l Effe
ct
Uptake
● High
Low
High uptakevillage
Low uptakevillage
0.00 0.03 0.06 0.09
Average marginal effect of one adopting neighbor on adoption
Effect
satisfaction
communication
contagion
●
●
●
●
Low uptake
High uptake
−0.01 0.00 0.01 0.02 0.03 0.04 0.05
Leader
Peer
Leader
Peer
Average Marginal Effect
sour
ce
●
●
●
●
●
●
●
●
●
●
Ethnic concentration (12)
Pct. strong ties (15)
Pct. leaders (15)
Degree (15)
Diffusion potential (15)
Population (15)
Religious concentration (15)
Pro−sociality − dictator (15)
Pro−sociality − public good (15)
Leadership concentration (14)
−5.0 −2.5 0.0 2.5
Standardized effect size
Var
iabl
e
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 26 / 29
Empirical implications
Findings
1 Variation across networks in the support of diffusion of goods withexternalities
There are peer effects, but not in all villages.
2 Discounting of positive signals (peers’ recommendations) when truthfulcommunication is not enforced
i is more likely to adopt if j is satisfied in high uptake villages
3 Experts will have a stronger peer effect than novices
Leaders exert more influence than citizens in high-uptake villages
4 Informal institutions should support adoption in high-uptake villages
Public goods games and concentrated leadership
●
●●
●
●
●●
−0.05
0.00
0.05
0.10
B(236)
H(102)
E(263)
C(159)
M(195)
G(163)
D(282)
K(204)
L(228)
P(191)
O(187)
I(168)
F(205)
N(223)
J(185)
Village
Ave
rage
Mar
gina
l Effe
ct
Uptake
● High
Low
High uptakevillage
Low uptakevillage
0.00 0.03 0.06 0.09
Average marginal effect of one adopting neighbor on adoption
Effect
satisfaction
communication
contagion
●
●
●
●
Low uptake
High uptake
−0.01 0.00 0.01 0.02 0.03 0.04 0.05
Leader
Peer
Leader
Peer
Average Marginal Effect
sour
ce
●
●
●
●
●
●
●
●
●
●
Ethnic concentration (12)
Pct. strong ties (15)
Pct. leaders (15)
Degree (15)
Diffusion potential (15)
Population (15)
Religious concentration (15)
Pro−sociality − dictator (15)
Pro−sociality − public good (15)
Leadership concentration (14)
−5.0 −2.5 0.0 2.5
Standardized effect size
Var
iabl
e
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 26 / 29
Empirical implications
Findings
1 Variation across networks in the support of diffusion of goods withexternalities
There are peer effects, but not in all villages.
2 Discounting of positive signals (peers’ recommendations) when truthfulcommunication is not enforced
i is more likely to adopt if j is satisfied in high uptake villages
3 Experts will have a stronger peer effect than novices
Leaders exert more influence than citizens in high-uptake villages
4 Informal institutions should support adoption in high-uptake villages
Public goods games and concentrated leadership
●
●●
●
●
●●
−0.05
0.00
0.05
0.10
B(236)
H(102)
E(263)
C(159)
M(195)
G(163)
D(282)
K(204)
L(228)
P(191)
O(187)
I(168)
F(205)
N(223)
J(185)
Village
Ave
rage
Mar
gina
l Effe
ct
Uptake
● High
Low
High uptakevillage
Low uptakevillage
0.00 0.03 0.06 0.09
Average marginal effect of one adopting neighbor on adoption
Effect
satisfaction
communication
contagion
●
●
●
●
Low uptake
High uptake
−0.01 0.00 0.01 0.02 0.03 0.04 0.05
Leader
Peer
Leader
Peer
Average Marginal Effect
sour
ce
●
●
●
●
●
●
●
●
●
●
Ethnic concentration (12)
Pct. strong ties (15)
Pct. leaders (15)
Degree (15)
Diffusion potential (15)
Population (15)
Religious concentration (15)
Pro−sociality − dictator (15)
Pro−sociality − public good (15)
Leadership concentration (14)
−5.0 −2.5 0.0 2.5
Standardized effect size
Var
iabl
e
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 26 / 29
Empirical implications
Findings
1 Variation across networks in the support of diffusion of goods withexternalities
There are peer effects, but not in all villages.
2 Discounting of positive signals (peers’ recommendations) when truthfulcommunication is not enforced
i is more likely to adopt if j is satisfied in high uptake villages
3 Experts will have a stronger peer effect than novices
Leaders exert more influence than citizens in high-uptake villages
4 Informal institutions should support adoption in high-uptake villages
Public goods games and concentrated leadership
●
●●
●
●
●●
−0.05
0.00
0.05
0.10
B(236)
H(102)
E(263)
C(159)
M(195)
G(163)
D(282)
K(204)
L(228)
P(191)
O(187)
I(168)
F(205)
N(223)
J(185)
Village
Ave
rage
Mar
gina
l Effe
ct
Uptake
● High
Low
High uptakevillage
Low uptakevillage
0.00 0.03 0.06 0.09
Average marginal effect of one adopting neighbor on adoption
Effect
satisfaction
communication
contagion
●
●
●
●
Low uptake
High uptake
−0.01 0.00 0.01 0.02 0.03 0.04 0.05
Leader
Peer
Leader
Peer
Average Marginal Effect
sour
ce
●
●
●
●
●
●
●
●
●
●
Ethnic concentration (12)
Pct. strong ties (15)
Pct. leaders (15)
Degree (15)
Diffusion potential (15)
Population (15)
Religious concentration (15)
Pro−sociality − dictator (15)
Pro−sociality − public good (15)
Leadership concentration (14)
−5.0 −2.5 0.0 2.5
Standardized effect size
Var
iabl
e
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 26 / 29
Empirical implications
Findings
1 Variation across networks in the support of diffusion of goods withexternalities
There are peer effects, but not in all villages.
2 Discounting of positive signals (peers’ recommendations) when truthfulcommunication is not enforced
i is more likely to adopt if j is satisfied in high uptake villages
3 Experts will have a stronger peer effect than novicesLeaders exert more influence than citizens in high-uptake villages
4 Informal institutions should support adoption in high-uptake villages
Public goods games and concentrated leadership
●
●●
●
●
●●
−0.05
0.00
0.05
0.10
B(236)
H(102)
E(263)
C(159)
M(195)
G(163)
D(282)
K(204)
L(228)
P(191)
O(187)
I(168)
F(205)
N(223)
J(185)
Village
Ave
rage
Mar
gina
l Effe
ct
Uptake
● High
Low
High uptakevillage
Low uptakevillage
0.00 0.03 0.06 0.09
Average marginal effect of one adopting neighbor on adoption
Effect
satisfaction
communication
contagion
●
●
●
●
Low uptake
High uptake
−0.01 0.00 0.01 0.02 0.03 0.04 0.05
Leader
Peer
Leader
Peer
Average Marginal Effect
sour
ce
●
●
●
●
●
●
●
●
●
●
Ethnic concentration (12)
Pct. strong ties (15)
Pct. leaders (15)
Degree (15)
Diffusion potential (15)
Population (15)
Religious concentration (15)
Pro−sociality − dictator (15)
Pro−sociality − public good (15)
Leadership concentration (14)
−5.0 −2.5 0.0 2.5
Standardized effect size
Var
iabl
e
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 26 / 29
Empirical implications
Findings
1 Variation across networks in the support of diffusion of goods withexternalities
There are peer effects, but not in all villages.
2 Discounting of positive signals (peers’ recommendations) when truthfulcommunication is not enforced
i is more likely to adopt if j is satisfied in high uptake villages
3 Experts will have a stronger peer effect than novicesLeaders exert more influence than citizens in high-uptake villages
4 Informal institutions should support adoption in high-uptake villages
Public goods games and concentrated leadership
●
●●
●
●
●●
−0.05
0.00
0.05
0.10
B(236)
H(102)
E(263)
C(159)
M(195)
G(163)
D(282)
K(204)
L(228)
P(191)
O(187)
I(168)
F(205)
N(223)
J(185)
Village
Ave
rage
Mar
gina
l Effe
ct
Uptake
● High
Low
High uptakevillage
Low uptakevillage
0.00 0.03 0.06 0.09
Average marginal effect of one adopting neighbor on adoption
Effect
satisfaction
communication
contagion
●
●
●
●
Low uptake
High uptake
−0.01 0.00 0.01 0.02 0.03 0.04 0.05
Leader
Peer
Leader
Peer
Average Marginal Effect
sour
ce
●
●
●
●
●
●
●
●
●
●
Ethnic concentration (12)
Pct. strong ties (15)
Pct. leaders (15)
Degree (15)
Diffusion potential (15)
Population (15)
Religious concentration (15)
Pro−sociality − dictator (15)
Pro−sociality − public good (15)
Leadership concentration (14)
−5.0 −2.5 0.0 2.5
Standardized effect size
Var
iabl
e
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 26 / 29
Empirical implications
Findings
1 Variation across networks in the support of diffusion of goods withexternalities
There are peer effects, but not in all villages.
2 Discounting of positive signals (peers’ recommendations) when truthfulcommunication is not enforced
i is more likely to adopt if j is satisfied in high uptake villages
3 Experts will have a stronger peer effect than novices
Leaders exert more influence than citizens in high-uptake villages
4 Informal institutions should support adoption in high-uptake villages
Public goods games and concentrated leadership
●
●●
●
●
●●
−0.05
0.00
0.05
0.10
B(236)
H(102)
E(263)
C(159)
M(195)
G(163)
D(282)
K(204)
L(228)
P(191)
O(187)
I(168)
F(205)
N(223)
J(185)
Village
Ave
rage
Mar
gina
l Effe
ct
Uptake
● High
Low
High uptakevillage
Low uptakevillage
0.00 0.03 0.06 0.09
Average marginal effect of one adopting neighbor on adoption
Effect
satisfaction
communication
contagion
●
●
●
●
Low uptake
High uptake
−0.01 0.00 0.01 0.02 0.03 0.04 0.05
Leader
Peer
Leader
Peer
Average Marginal Effect
sour
ce
●
●
●
●
●
●
●
●
●
●
Ethnic concentration (12)
Pct. strong ties (15)
Pct. leaders (15)
Degree (15)
Diffusion potential (15)
Population (15)
Religious concentration (15)
Pro−sociality − dictator (15)
Pro−sociality − public good (15)
Leadership concentration (14)
−5.0 −2.5 0.0 2.5
Standardized effect size
Var
iabl
e
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 26 / 29
Empirical implications
Findings
1 Variation across networks in the support of diffusion of goods withexternalities
There are peer effects, but not in all villages.
2 Discounting of positive signals (peers’ recommendations) when truthfulcommunication is not enforced
i is more likely to adopt if j is satisfied in high uptake villages
3 Experts will have a stronger peer effect than novicesLeaders exert more influence than citizens in high-uptake villages
4 Informal institutions should support adoption in high-uptake villagesPublic goods games and concentrated leadership
●
●●
●
●
●●
−0.05
0.00
0.05
0.10
B(236)
H(102)
E(263)
C(159)
M(195)
G(163)
D(282)
K(204)
L(228)
P(191)
O(187)
I(168)
F(205)
N(223)
J(185)
Village
Ave
rage
Mar
gina
l Effe
ct
Uptake
● High
Low
High uptakevillage
Low uptakevillage
0.00 0.03 0.06 0.09
Average marginal effect of one adopting neighbor on adoption
Effect
satisfaction
communication
contagion
●
●
●
●
Low uptake
High uptake
−0.01 0.00 0.01 0.02 0.03 0.04 0.05
Leader
Peer
Leader
Peer
Average Marginal Effect
sour
ce
●
●
●
●
●
●
●
●
●
●
Ethnic concentration (12)
Pct. strong ties (15)
Pct. leaders (15)
Degree (15)
Diffusion potential (15)
Population (15)
Religious concentration (15)
Pro−sociality − dictator (15)
Pro−sociality − public good (15)
Leadership concentration (14)
−5.0 −2.5 0.0 2.5
Standardized effect size
Var
iabl
e
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 26 / 29
Conclusions
Conclusion
1 We qualify a long-standing argument: “peer effects are ubiquitous in theprocess of technology adoption.”
2 For technologies with strong externalities, there are no peer effects ifcommunities do not manage to enforce truthful communication.
3 This may explain variation, and overall low rates of adoption of ICTs forpolitical communication.
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 27 / 29
Appendix
Example messages
Not relevant:“Hi ubridge”“We are for election”
Relevant:“I greet you all, but our major problem is sickness”“The tobbacco farmers are misserable how can Ubridge help them?”
Actionable:“The Only Borehole in Ogboa Village is broken”“NURSES DONT ATTEND PATIENTS DURING SAT AND sun in Opia HealthCentre”
Grossman (UPenn) Peer effects and externalities in technology adoption December 15, 2017 28 / 29
top related